We Keep Coding

Python gensim (TfidfModel): How is the Tf-Idf computed?

1. For the below test text,
test=['test test', 'test toy']
the tf-idf score [without normalisation (smartirs: 'ntn')] is
[['test', 1.17]]
[['test', 0.58], ['toy', 1.58]]
This doesn't seem to tally with what I get via direct computation of
tfidf (w, d) = tf x idf
where idf(term)=log (total number of documents / number of documents containing term)
tf = number of instances of word in d document / total number of words of d document
Eg
doc 1: 'test test'
for "test" word
tf= 1
idf= log(2/2) = 0
tf-idf = 0
Can someone show me the computation using my above test text?
2) When I change to cosine normalisation (smartirs:'ntc'), I get
[['test', 1.0]]
[['test', 0.35], ['toy', 0.94]]
Can someone show me the computation too?
Thank you
import gensim
from gensim import corpora
from gensim import models
import numpy as np
from gensim.utils import simple_preprocess
test=['test test', 'test toy']
texts = [simple_preprocess(doc) for doc in test]
mydict= corpora.Dictionary(texts)
mycorpus = [mydict.doc2bow(doc, allow_update=True) for doc in texts]
tfidf = models.TfidfModel(mycorpus, smartirs='ntn')
for doc in tfidf[mycorpus]:
print([[mydict[id], np.around(freq, decimals=2)] for id, freq in doc])

If you care to know the details of the implementation of the model.TfidfModel you can check them directly in the GitHub repository for gensim. The particular calculation scheme corresponding to smartirs='ntn' is described on the Wikipedia page for SMART Information Retrieval System and the exact calculations are different than the ones you use hence the difference in the results.
E.g. the particular discrepancy you are referring to:
idf= log(2/2) = 0
should actually be log2(N+1/n_k):
idf= log(2/1) = 1
I suggest that you review both the implementation and the mentioned page so that to ensure your manual checks follow the implementation for the chosen smartirs flags.

SciSpacy equivalent of Gensim's functions/parameters

With Gensim, there are three functions I use regularly, for example this one:
model = gensim.models.Word2Vec(corpus,size=100,min_count=5)
The output from gensim, but I cannot understand how to set the size and min_count parameters in the equivalent SciSpacy command of:
model = spacy.load('en_core_web_md')
(The output is a model of embeddings (too big to add here))).
This is another command I regularly use:
model.most_similar(positive=['car'])
and this is the output from gensim/Expected output from SciSpacy:
[('vehicle', 0.7857330441474915),
('motorbike', 0.7572781443595886),
('train', 0.7457204461097717),
('honda', 0.7383008003234863),
('volkswagen', 0.7298516035079956),
('mini', 0.7158907651901245),
('drive', 0.7093928456306458),
('driving', 0.7084407806396484),
('road', 0.7001082897186279),
('traffic', 0.6991947889328003)]
This is the third command I regularly use:
print(model.wv['car'])
Output from Gensim/Expected output from SciSpacy (in reality this vector is length 100):
[ 1.0942473 2.5680697 -0.43163642 -1.171171 1.8553845 -0.3164575
1.3645878 -0.5003705 2.912658 3.099512 2.0184739 -1.2413547
0.9156444 -0.08406237 -2.2248871 2.0038593 0.8751471 0.8953876
0.2207374 -0.157277 -1.4984075 0.49289042 -0.01171476 -0.57937795...]
Could someone show me the equivalent commands for SciSpacy? For example, for 'gensim.models.Word2Vec' I can't find how to specify the length of the vectors (size parameter), or the minimum number of times the word should be in the corpus (min_count) in SciSpacy (e.g. I looked here and here), but I'm not sure if I'm missing them?

A possible way to achieve your goal would be to:
parse you documents via nlp.pipe
collect all the words and pairwise similarities
process similarities to get the desired results
Let's prepare some data:
import spacy
nlp = spacy.load("en_core_web_md", disable = ['ner', 'tagger', 'parser'])
Then, to get a vector, like in model.wv['car'] one would do:
nlp("car").vector
To get most similar words like model.most_similar(positive=['car']) let's process the corpus:
corpus = ["This is a sentence about cars. This a sentence aboout train"
, "And this is a sentence about a bike"]
docs = nlp.pipe(corpus)
tokens = []
tokens_orth = []
for doc in docs:
for tok in doc:
if tok.orth_ not in tokens_orth:
tokens.append(tok)
tokens_orth.append(tok.orth_)
sims = np.zeros((len(tokens),len(tokens)))
for i, tok in enumerate(tokens):
sims[i] = [tok.similarity(tok_) for tok_ in tokens]
Then to retrieve top=3 most similar words:
def most_similar(word, tokens_orth = tokens_orth, sims=sims, top=3):
tokens_orth = np.array(tokens_orth)
id_word = np.where(tokens_orth == word)[0][0]
sim = sims[id_word]
id_ms = np.argsort(sim)[:-top-1:-1]
return list(zip(tokens_orth[id_ms], sim[id_ms]))
most_similar("This")
[('this', 1.0000001192092896), ('This', 1.0), ('is', 0.5970357656478882)]
PS
I have also noticed you asked for specification of dimension and frequency. Embedding length is fixed at the time the model is initialized, so it can't be changed after that. You can start from a blank model if you wish so, and feed embeddings you're comfortable with. As for the frequency, it's doable, via counting all the words and throwing away anything that is below desired threshold. But again, underlying embeddings will be from a not filtered text. SpaCy is different from Gensim in that it uses readily available embeddings whereas Gensim trains them.

Re-calculate similarity matrix given new documents

I'm running an experiment that include text documents that I need to calculate the (cosine) similarity matrix between all of them (to use for another calculation). For that I use sklearn's TfidfVectorizer:
corpus = [doc1, doc2, doc3, doc4]
vect = TfidfVectorizer(min_df=1, stop_words="english", use_idf=False)
tfidf = vect.fit_transform(corpus)
similarities = tfidf * tfidf.T
pairwise_similarity_matrix = similarities.A
The problem is that with each iteration of my experiment I discover new documents that I need to add to my similarity matrix, and given the number of documents I'm working with (tens of thousands and more) - it is very time consuming.
I wish to find a way to calculate only the similarities between the new batch of documents and the existing ones, without computing it all again one the entire data set.
Note that I'm using a term-frequency (tf) representation, without using inverse-document-frequency (idf), so in theory I don't need to re-calculate the whole matrix each time.

OK, I got it.
The idea is, as I said, to calculate the similarity only between the new batch of files and the existing ones, which their similarity is unchanged. The problem is to keep the TfidfVectorizer's vocabulary updated with the newly seen terms.
The solution has 2 steps:
Update the vocabulary and the tf matrices.
Matrix multiplications and stacking.
Here's the whole script - we first got the original corpus and the trained and calculated objects and matrices:
corpus = [doc1, doc2, doc3]
# Build for the first time:
vect = TfidfVectorizer(min_df=1, stop_words="english", use_idf=False)
tf_matrix = vect.fit_transform(corpus)
similarities = tf_matrix * tf_matrix.T
similarities_matrix = similarities.A # just for printing
Now, given new documents:
new_docs_corpus = [docx, docy, docz] # New documents
# Building new vectorizer to create the parsed vocabulary of the new documents:
new_vect = TfidfVectorizer(min_df=1, stop_words="english", use_idf=False)
new_vect.fit(new_docs_corpus)
# Merging old and new vocabs:
new_terms_count = 0
for k, v in new_vect.vocabulary_.items():
if k in vect.vocabulary_.keys():
continue
vect.vocabulary_[k] = np.int64(len(vect.vocabulary_)) # important not to assign a simple int
new_terms_count = new_terms_count + 1
new_vect.vocabulary_ = vect.vocabulary_
# Build new docs represantation using the merged vocabulary:
new_tf_matrix = new_vect.transform(new_docs_corpus)
new_similarities = new_tf_matrix * new_tf_matrix.T
# Get the old tf-matrix with the same dimentions:
if new_terms_count:
zero_matrix = csr_matrix((tfidf.shape[0],new_terms_count))
tf_matrix = hstack([tf_matrix, zero_matrix])
# tf_matrix = vect.transform(corpus) # Instead, we just append 0's for the new terms and stack the tf_matrix over the new one, to save time
cross_similarities = new_tf_matrix * tf_matrix.T # Calculate cross-similarities
tf_matrix = vstack([tf_matrix, new_tfidf])
# Stack it all together:
similarities = vstack([hstack([similarities, cross_similarities.T]), hstack([cross_similarities, new_similarities])])
similarities_matrix = similarities.A
# Updating the corpus with the new documents:
corpus = corpus + new_docs_corpus
We can check this by comparing the calculated similarities_matrix we got, with the one we get when we train a TfidfVectorizer on the joint corpus: corpus + new_docs_corpus.
As discussed in the the comments, we can do all that only because we are not using the idf (inverse-document-frequency) element, that will change the representation of existing documents given new ones.

How to make percentage bar chart of topics from topic modeling?

I have been racking my brain on this for a week.
I want to
run NMF topic modeling
Assign each document a topic by looking at the maximum of weights,
Graph this distribution as a % bar chart using matplot. (I.e: Topics on the X axis, and % documents that are that topic on the y axis. )
Here is some toy data and completing steps 1 and 2:
from sklearn.feature_extraction.text import TfidfVectorizer
from sklearn.decomposition import NMF
import pandas as pd
# Get data
data = {
"Documents": ["I am a document",
"And me too",
"The cat is big",
"The dog is big"
"My headphones are large",
"My monitor has rabies",
"My headphones are loud"
"The street is loud "]
}
df = pd.DataFrame(data)
# Fit a TFIDF vectorizer
tfidf_vectorizer = TfidfVectorizer()
tfidf = tfidf_vectorizer.fit_transform(df['Documents'])
# Run NMF
nmf_model = NMF(n_components=4, random_state=1).fit(tfidf)
# Weights
W = nmf_model.transform(tfidf)
# Topics
H= nmf_model.components_
Now here is how I can assign a document to a topcic:
# Will return document topics as list like [1, 4, 1...] to
# represent that the first document is topic 1, the second 4, and so on.
topics = pd.DataFrame(W).idxmax(axis=1, skipna=True).tolist()
Alright now I should be able to get what I want with these two structures but I am at a loss.

looks like a use case for a Counter().
I'd write something like this:
from collections import Counter
mylist = [1,1,1,1,2,2,3,1,1,2,3,1,1,1]
mycount = Counter(mylist)
for key,value in mycount.items():
print(key,value)
This outputs your topics in the following structure:
1 9
2 3
3 2
One thing to note for Latent dirichlet / non negative matrix is that the entire point is a sentence is composed of multiple topics. Maxing the weight to assign each to a single topic may defeat the purpose. You also may want to consider how to deal with nonsense sentences, as your algorithm will auto assign them to a topic currently.

IIUC, you want to draw a bar char, so do not change topics into list:
topics = pd.DataFrame(W).idxmax(axis=1, skipna=True)
plt.bar(x=topics.index, height=topics.mul(100)/topics.sum())
plt.show()
gives:

How to call a corpus file in python?

I am currently working on gensim doc2vec model to implement sentence similarity.
I came across this sample code by William Bert where he has mentioned that to train this model I need to provide my own background corpus. The code is copied below for convenience:
import logging, sys, pprint
logging.basicConfig(stream=sys.stdout, level=logging.INFO)
### Generating a training/background corpus from your own source of documents
from gensim.corpora import TextCorpus, MmCorpus, Dictionary
# gensim docs: "Provide a filename or a file-like object as input and TextCorpus will be initialized with a
# dictionary in `self.dictionary`and will support the `iter` corpus method. For other kinds of corpora, you only
# need to override `get_texts` and provide your own implementation."
background_corpus = TextCorpus(input=YOUR_CORPUS)
# Important -- save the dictionary generated by the corpus, or future operations will not be able to map results
# back to original words.
background_corpus.dictionary.save(
"my_dict.dict")
MmCorpus.serialize("background_corpus.mm",
background_corpus) # Uses numpy to persist wiki corpus in Matrix Market format. File will be several GBs.
### Generating a large training/background corpus using Wikipedia
from gensim.corpora import WikiCorpus, wikicorpus
articles = "enwiki-latest-pages-articles.xml.bz2" # available from http://en.wikipedia.org/wiki/Wikipedia:Database_download
# This will take many hours! Output is Wikipedia in bucket-of-words (BOW) sparse matrix.
wiki_corpus = WikiCorpus(articles)
wiki_corpus.dictionary.save("wiki_dict.dict")
MmCorpus.serialize("wiki_corpus.mm", wiki_corpus) # File will be several GBs.
### Working with persisted corpus and dictionary
bow_corpus = MmCorpus("wiki_corpus.mm") # Revive a corpus
dictionary = Dictionary.load("wiki_dict.dict") # Load a dictionary
### Transformations among vector spaces
from gensim.models import LsiModel, LogEntropyModel
logent_transformation = LogEntropyModel(wiki_corpus,
id2word=dictionary) # Log Entropy weights frequencies of all document features in the corpus
tokenize_func = wikicorpus.tokenize # The tokenizer used to create the Wikipedia corpus
document = "Some text to be transformed."
# First, tokenize document using the same tokenization as was used on the background corpus, and then convert it to
# BOW representation using the dictionary created when generating the background corpus.
bow_document = dictionary.doc2bow(tokenize_func(
document))
# converts a single document to log entropy representation. document must be in the same vector space as corpus.
logent_document = logent_transformation[[
bow_document]]
# Transform arbitrary documents by getting them into the same BOW vector space created by your training corpus
documents = ["Some iterable", "containing multiple", "documents", "..."]
bow_documents = (dictionary.doc2bow(
tokenize_func(document)) for document in documents) # use a generator expression because...
logent_documents = logent_transformation[
bow_documents] # ...transformation is done during iteration of documents using generators, so this uses constant memory
### Chained transformations
# This builds a new corpus from iterating over documents of bow_corpus as transformed to log entropy representation.
# Will also take many hours if bow_corpus is the Wikipedia corpus created above.
logent_corpus = MmCorpus(corpus=logent_transformation[bow_corpus])
# Creates LSI transformation model from log entropy corpus representation. Takes several hours with Wikipedia corpus.
lsi_transformation = LsiModel(corpus=logent_corpus, id2word=dictionary,
num_features=400)
# Alternative way of performing same operation as above, but with implicit chaining
# lsi_transformation = LsiModel(corpus=logent_transformation[bow_corpus], id2word=dictionary,
# num_features=400)
# Can persist transformation models, too.
logent_transformation.save("logent.model")
lsi_transformation.save("lsi.model")
### Similarities (the best part)
from gensim.similarities import Similarity
# This index corpus consists of what you want to compare future queries against
index_documents = ["A bear walked in the dark forest.",
"Tall trees have many more leaves than short bushes.",
"A starship may someday travel across vast reaches of space to other stars.",
"Difference is the concept of how two or more entities are not the same."]
# A corpus can be anything, as long as iterating over it produces a representation of the corpus documents as vectors.
corpus = (dictionary.doc2bow(tokenize_func(document)) for document in index_documents)
index = Similarity(corpus=lsi_transformation[logent_transformation[corpus]], num_features=400, output_prefix="shard")
print "Index corpus:"
pprint.pprint(documents)
print "Similarities of index corpus documents to one another:"
pprint.pprint([s for s in index])
query = "In the face of ambiguity, refuse the temptation to guess."
sims_to_query = index[lsi_transformation[logent_transformation[dictionary.doc2bow(tokenize_func(query))]]]
print "Similarities of index corpus documents to '%s'" % query
pprint.pprint(sims_to_query)
best_score = max(sims_to_query)
index = sims_to_query.tolist().index(best_score)
most_similar_doc = documents[index]
print "The document most similar to the query is '%s' with a score of %.2f." % (most_similar_doc, best_score)
Where and how should I provide my own corpus in the code?
Thanks in advance for your help.